This page describes the impact that hardware characteristics can have on performance of CSI programs.

On this page:

Overview

Analysis, design, and results-recovery times are dependent upon the following hardware characteristics:

• Available memory
• Input/output (I/O)
• Processor speed
• Number of cores

Depending on the type of analysis, some hardware characteristics may have greater impact on analysis and results-recovery speed than others.

Hardware requirements for analysis types

This table approximates the hardware demand for various analysis types. I/O intensive analyses tend to benefit from both increased disk speed and additional RAM. Fast SSDs directly connected to the PCI bus (PCIe) are recommended over regular hard drives connected to a SATA interface or slower SSDs. External hard drives are not recommended, and network drives should not be used. The operating system uses RAM for file caching. The most cost-effective approach to improving software performance is to increase the amount of RAM on 64-bit operating systems, and to run analysis out-of-process.

Additional Information

(1) Nonlinear modal time history is parallelized for state determination and results recovery.

(2) Design and calculation of story drifts & shears in ETABS are parallelized hence will benefit from multiple cores.

(3) In general, Intel Core series CPUs (i5/i7/i9) perform better compared to Intel Xeons (and AMD CPUs).

(4) In general, CPUs with higher clock speed with less number cores perform better compared to CPUs with lower clock speeds with more number cores.

(5) The number of processors that can run load cases in parallel in a given session is limited to eight. This prevents excessive competition for disk resources and allows for more internal parallelization of certain algorithmic processes since each load case, run serially or in parallel, can use up to all physical cores on the machine. It is still recommended to limit the number of parallel load cases to the number of physical cores on the machine if less than eight.

Which CPUs are best for analysis? (updated May 2021)

Considerations:

• Recent mainstream & performance Intel desktop/workstation CPUs (Core i5, i7, i9) are expected to perform well.
• Recent mainstream & performance AMD Ryzen CPUs based on Zen 2 & Zen 3 architecture are expected to perform similar to or better compared to Intel CPUs with similar specs (base & turbo clocks, Core i5 for mainstream and Core i7/i9 for performance, year of release)
• Recent Intel and AMD server CPUs with many cores but low base & turbo clocks (e.g. AMD Epyc, Intel Xeon) do not perform well compared to desktop/workstation CPUs (e.g.) with less number of cores with high base & turbo clocks.
  • High base and turbo clock speeds are critical for best analysis performance. Server CPUs usually have lower base clock speeds and cannot turbo as high as desktop CPUs when multiple cores are idle. This eats away the performance gain from high core count, and results in overall longer run times.
• Older (>5 years) Intel & AMD desktop/workstation CPUs perform considerably worse compared to newer offerings from Intel & AMD with similar base clock speeds:
  • Newer CPUs can perform more instructions per clock cycle in addition to much higher base and turbo clocks, resulting in much faster run times.
• Desktop CPUs compared to similarly spec’d laptop CPUs perform better due to laptop CPUs’ thermal constraints preventing them from reaching turbo frequencies early & for long.

Conclusion:

• Recent Intel Core i5/i7/i9 and/or AMD Ryzen 5/7/9 series with 6-8 cores and high (> 3.50 GHz) base and turbo (> 4.00 GHz) clocks are recommended for best analysis performance. CPUs with more than 8 cores may be useful when running many load cases in parallel.
• Laptop CPUs with low turbo clocks & thermal constraints and server CPUs with many cores with low base & turbo clocks are not recommended.

See Also

• Minimum and Recommended System Requirements article

• Reducing computational time for influence-based moving load analysis article

• Solver section

• Parallel processing article